FSBKT: Federated Structural and Behavioral Prototype Learning for Privacy-Preserving Knowledge Tracing

The implementation of the paper FSBKT: Federated Structural and Behavioral Prototype Learning for Privacy-Preserving Knowledge Tracing.

Abstract

Knowledge tracing aims to model students' evolving knowledge states and predict their future performance based on historical learning interactions. However, real-world educational data are typically distributed across schools with substantially different curricula, exercise structures, and student behavior patterns. Meanwhile, privacy constraints often prevent institutions from directly sharing student records.

To address these challenges, we propose FSBKT, a federated knowledge tracing framework based on structural and behavioral prototype learning. Each client extracts a structural prototype from its local exercise-oriented knowledge graph using a Graph Attention Network and a behavioral prototype from student interaction sequences using a GRU encoder. Local differential privacy is further applied to behavioral prototypes before transmission. On the server side, Mixture-of-Prototypes aggregation with Exponential Moving Average produces stable prototype references, while Similarity-based Personalized Aggregation computes client-specific model parameters through similarity-weighted peer aggregation. Experiments on multiple real-world knowledge tracing benchmarks demonstrate that FSBKT improves predictive performance under highly non-IID settings while maintaining favorable privacy–utility trade-offs.

Overall Architecture

Requirements

• Python 3.8+
• PyTorch 1.12+
• scikit-learn
• NumPy

Usage

Configure the dataset paths and model settings in:

KnowledgeTracing/Constant.py

Then run the main training script:

python main.py

For privacy evaluation, including membership inference attacks (MIA) and re-identification attacks (ReID), run:

python mia+reid.py

Citation

If you find this paper and repository useful for your research, please consider citing:

@inproceedings{yao2026fsbkt,
  author    = {Yao, Yumo and Yu, Hai-Tao and Chen, Jiankang},
  title     = {{FSBKT}: Federated Structural and Behavioral Prototype Learning for Privacy-Preserving Knowledge Tracing},
  booktitle = {2026 IEEE International Conference on Systems, Man, and Cybernetics (SMC)},
  year      = {2026},
  note      = {Accepted}
}

The citation information will be updated after the paper is published in IEEE Xplore.